Simulation Dataset for Deep-Learning Mitigation of Foregrounds and Beam Effects in 21-cm Intensity Mapping

This dataset contains simulated 21cm brightness temperature cubes for testing foreground removal, beam modelling, and 21cm signal reconstruction, employing the CRIME code. Four versions of the data are provided: pure HI, HI with astrophysical foregrounds, and HI+foregrounds convolved with either Gaussian or cosine-shaped primary beams. The included foreground components follow standard radio-astronomy sky models and contain: Galactic synchrotron radiation, Galactic and extragalactic free–free emissions, and extragalactic point sources. Each cube has shape (192, 64, 64, 64), corresponding to 192 slices along the line of sight, a spatial resolution of 64×64 pixels, and 64 frequency channels, we consider a frequency range spanning from 1100 to 1164 MHz, which corresponds to a redshift range between 0.29 and 0.22. The dataset is intended for testing 21cm reconstruction methods, machine learning models, and chromatic beam analyses. If you use this dataset, please cite both the dataset DOI and the related research works listed below: – Deep-learning mitigation of foregrounds and beam effects in 21-cm intensity mapping using hybrid frequency differencing and PCA, arXiv: https://arxiv.org/abs/2511.15072– 21-cm foreground removal using AI and frequency-difference technique, arXiv: https://arxiv.org/abs/2310.06518